Article Text

Download PDFPDF

Research paper
Increased mortality persists in an adult drug-resistant epilepsy prevalence cohort
  1. Brian Callaghan1,
  2. Hyunmi Choi2,
  3. Malka Schlesinger2,
  4. William Rodemer3,
  5. John Pollard3,
  6. Dale C Hesdorffer2,
  7. W Allen Hauser2,
  8. Jacqueline French4
  1. 1Department of Neurology, University of Michigan, Ann Arbor, Michigan, USA
  2. 2Columbia University, New York, New York, USA
  3. 3Department of Neurology, University of Pennsylvania, Philadelphia, Pennsylvania, USA
  4. 4New York University, New York, New York, USA
  1. Correspondence to Dr Brian Callaghan, Department of Neurology, University of Michigan, 109 Zina Pitcher Place, 4021 BSRB, Ann Arbor, MI 48104, USA; bcallagh{at}


Objective To investigate the cumulative probability of death and the standardised mortality ratio (SMR) in an adult drug-resistant epilepsy (DRE) population.

Methods In two separate centres during 2003–2006, we identified a total of 433 patients with DRE defined as at least one seizure per month and failure of at least two antiepileptic drugs. These patients were subsequently followed for a total follow-up of 6 years. We examined the cumulative probability of death, using Kaplan-Meier methodology, and the SMR based on mortality data from the Social Security Death Index. Clinical predictors of death were evaluated using Cox regression analysis.

Results The cumulative probability of death was 8.7% (95% CI 6.2% to 12.1%) at 6 years. The overall SMR was 2.4 (95% CI 1.7 to 3.3). It was 3.1; 95% CI 2.0 to 4.6 in subjects with remote or progressive aetiology and 1.7; 95% CI 0.8 to 2.8 in subjects with unknown aetiology. The SMR was significantly increased in those with a known remote aetiology (2.5; 95% CI (1.4 to 3.8)). Older age at enrolment and symptomatic generalised epilepsy syndrome were significant predictors of death.

Discussion Mortality is increased in this drug-resistant population; largely driven by those with a known epilepsy aetiology. The increased mortality remains even after exclusion of those with a progressive aetiology. Previous studies of incident epilepsy cohorts revealed increased mortality that declines to near-normal levels after the first several years, but in our DRE cohort, mortality remains elevated despite a median duration of epilepsy of 25 years at study entry.

  • Epilepsy
  • Neuroepidemiology
  • Anticonvulsants
View Full Text

Statistics from


Patients with epilepsy have an increased mortality in many studies.1–11 Death in these patients can be due to underlying diseases that can cause epilepsy, death related to epilepsy per se, and death that is unrelated to the epilepsy diagnosis. Deaths due to underlying diseases include central nervous system (CNS) neoplasms, cerebrovascular disease, CNS infection, and neurodegenerative disorders can lead to premature mortality. Deaths related to the epilepsy condition include sudden unexplained death in epilepsy (SUDEP), status epilepticus, accidents and suicide.12 Deaths unrelated to epilepsy include non-CNS cancer, heart disease, pneumonia and other causes that are typical in the general population.

To date, many studies in population-based incident cohorts have addressed mortality in patients with epilepsy, and uniformly find increased mortality with standardised mortality ratios (SMR) ranging from 1.6 to 6.4.1–3 5–7 9–11 These studies differed on whether there was a significantly increased mortality in epilepsy of unknown cause with two of five studies finding an increased mortality compared to that expected in the general population.1 ,3 However, all these studies revealed that patients with epilepsy of known aetiology had a significant increased SMR (range from 2.2 to 6.5).1–3 5–7 9–11

Despite the extensive literature on mortality in unselected populations with epilepsy, less has been published on deaths in patients with drug-resistant epilepsy (DRE) (table 1). Three groups have shown that mortality is increased in prevalent cohorts with chronic epilepsy.8 ,13 ,14 These studies included patients with a wide range of seizure frequencies and number of antiepileptic drug (AED) failures. Other studies have focused on specific patient groups including those participating in clinical trials, institutionalised, or referred for presurgical evaluation.13 ,15–17 None of these studies required patients to meet a stricter definition of DRE.

Table 1

Studies investigating mortality in chronic epilepsy

We have previously identified two retrospective prevalence cohorts of patients with DRE and frequent seizures to look at seizure outcomes over time. This cohort included patients who had failed two or more appropriate antiepileptic regimens, and were having at least monthly seizures at their initial ascertainment. The purpose of the present study was to ascertain the case fatality, cumulative probability of death over 6 years, the SMR and determinants of mortality.


We identified two retrospective cohorts of prevalent DRE patients from large epilepsy centres.

University of Pennsylvania Epilepsy Center cohort: In 2003, we identified 246 patients who met the following definition of DRE: (1) greater than or equal to one seizure per month for the 3 months prior to inclusion; (2) failed greater than or equal to two AEDs prior to the index date; (3) were seen in 2000. The index date is the first date the patient was seen in 2000.18

The treating epileptologist's notes were used to determine drug failure.

All patients were required to have at least one follow-up visit during the first year.

Columbia University Epilepsy Center cohort: Between 2005 and 2006, we identified 187 patients, meeting the same definition of DRE, who were seen in 2001. The index date was defined as the first date the patient was seen in 2001.19

Population demographics and epilepsy characteristics were not significantly different between these two cohorts and, therefore, they were combined. Both cohorts were followed for the outcome of death for a total follow-up period of 6 years. At enrolment, we documented routine demographics, aetiology of epilepsy, epilepsy syndrome, age of onset of epilepsy, age of first meeting criteria for DRE, and number of AEDs failed. Since cohorts were identified in 2003 and 2005/2006, epilepsy classifications were made prior to the current proposed international league against epilepsy (ILAE) classification scheme. Follow-up methods included chart review and a search of the Social Security Death Index (SSDI) for all patients. We included all deaths captured through the year 2005 in the University of Pennsylvania cohort, and through 2006 in the Columbia University cohort. Among deceased patients, we requested death certificates to determine cause of death. Cause of death was also informed by medical chart abstraction and autopsy reports when available. Patients were noted to be lost to follow-up if there were no further interactions at the epilepsy centre in the medical record. The University of Pennsylvania's and Columbia University's Institutional Review Boards approved this study.

Statistical analysis

Case fatality was calculated as the percentage of people with DRE who died during the study period. Death incidence rates were calculated as the number of deaths divided by the number of person-years of follow up. Kaplan–Meier methodology was used to determine the cumulative probability of death. Patients were censored if they were lost to follow-up since the SSDI does not capture all deaths.20 ,21 These analyses were repeated in those with and without a known aetiology of epilepsy. For those with a known aetiology, these analyses were also performed in those with epilepsy attributed to a static lesion (eg, stroke, traumatic brain injury) and in those with a progressive cause of epilepsy (eg, a brain tumour, degenerative brain disease). The SMR was calculated by comparing observed to expected deaths using 2005 population mortality data from the National Vital Statistics System database. Cox proportional hazards regression was used to examine predictors of mortality. Factors examined included age at index, age of onset, gender, aetiology of epilepsy, epilepsy syndrome and number of AEDs failed. Complete case analyses were used (6 patients had missing age of onset).


The combined DRE population consisted of 357 (82.5%) patients with focal epilepsy, 46 (10.6%) with ‘symptomatic generalised epilepsy’ (so designated prior to the 2010 new ILAE classification22), 25 (5.8%) with genetic generalised epilepsy, and 5 (1.2%) with another epilepsy syndrome (table 2). The median age at the index date was 40 years (IQR=31–49) with a range of 12–83 years. Fifty-six per cent of the population was female, and the median duration of epilepsy was 25 years (IQR=14–36). Fifty per cent of the population had a known epilepsy aetiology. Of those with a known aetiology, 95%, had a remote static aetiology (eg, history of encephalitis or traumatic brain injury) and 5% had an enduring progressive aetiology (eg, brain tumour or neurodegenerative disease). Demographics did not differ between the two Epilepsy Center cohorts.

Table 2

Demographics and clinical characteristics of the entire adult drug-resistant epilepsy cohort and the deceased

The median age at the index date was greater for deceased patients than for patients still alive at the end of follow-up (46 vs 40 years; p<0.001). There was no difference in gender or median duration of epilepsy between these two groups.


Thirty-three of the 433 patients died during the course of follow-up (median 6.0 years, IQR=4.9–7.5) for a case fatality of 7.6%. In patients with a known aetiology of epilepsy the case fatality was 10.1%, whereas in those with an unknown aetiology the case fatality was 5.1% (p=0.05). Not surprisingly, among those with a known aetiology, the case fatality was 45.5% for progressive aetiology, and 8.2% for a remote aetiology (p<0.001). The death incidence rate was 13.4/1000 patient-years in the entire cohort, 17.9/1000 patient-years in those with known aetiology of epilepsy, and 8.9/1000 patient-years in those with unknown aetiology. In the first 2 years after the index date, 10 patients died and 29 were lost to follow-up. From years 2 to 4, an additional 12 patients died and 28 were lost to follow-up, and during years 4–6, 11 patients died with 27 were lost to follow-up.

Causes of death are displayed in table 3. Ten of the 33 patients who died had known active cancer at the time of the index date, including five with primary brain neoplasms. Two other patients had a known neurodegenerative condition. Deaths that were definitely or potentially epilepsy-related occurred in 10 patients: 5 SUDEP (3 with unknown and 2 with known aetiology), 3 status epilepticus, 1 accident, and 1 suicide. Seven deaths were associated with the underlying cause of the epilepsy (5 brain tumours and 2 strokes) with the remaining 16 deaths unrelated to the epilepsy or the underlying aetiology. In the entire cohort, the epileptologist did not know that 11 of 33 patients were deceased, (ie, the last chart note made no mention of death, and another appointment was scheduled) although three of these patients had either a primary or secondary brain tumour. Furthermore, 12% of those presumed to be lost to follow-up based on the medical chart, were found to be deceased.

Table 3

Causes of death and comorbidities at index

Kaplan–Meier analysis

The 6-year cumulative probability of death was 8.7% in the entire cohort (figure 1): 11.5% in those with known aetiology, and 5.8% in those with unknown aetiology (p=0.06) (figure 2). The 6-year cumulative probability of death was 45.5% in those with a known progressive cause and 9.5% in those with a known remote cause (p<0.0001) (figure 3).

Figure 1

Kaplan–Meier curve of time to death in the entire cohort.

Figure 2

Kaplan–Meier curves of time to death in those with known and unknown epilepsy aetiology.

Figure 3

Kaplan–Meier curves of time to death in those with known progressive and known non-progressive epilepsy aetiology.

Standardised mortality ratio

The SMR for the entire cohort was 2.4 (95% CI 1.7 to 3.3). For those with a known aetiology of epilepsy the SMR was 3.1 (2.0 to 4.6), and for those with an unknown aetiology, it was 1.7 (0.8 to 2.8). In those with a known aetiology, the SMR was 42.5 (13.4 to 87.9) in those with a progressive cause, and 2.5 (1.4 to 3.8) in those with a remote cause.

Cox proportional hazards regression

In adjusted analysis, factors associated with a statistically significant increased risk of death were older age at index HR (1.05 per year; 95% CI 1.02 to 1.07) and symptomatic generalised epilepsy syndrome (HR 4.28 (1.13 to 16.19); table 4). There was a trend for known aetiology of epilepsy to be associated with an increased risk of mortality (HR 2.17 (0.99 to 4.78)). Age of onset, gender, and number of AEDs failed were not significant predictors of mortality.

Table 4

Cox regression analysis


Most previous studies evaluating mortality in incident epilepsy have focused on the SMR.1–3 5–7 9–11 There has been a consensus that patients with epilepsy, particularly those with a known aetiology, have an increased number of observed deaths compared to the number expected in the general population, but a similar increase in mortality is observed regardless of aetiology.1–3 5–7 9–11 In these studies, SMR decreases over time, and for most incident cohorts with long-term follow-up, mortality is little changed from that expected 5–10 years following diagnosis. We report on mortality in the largest DRE cohort reported to date that meets a strict definition of drug resistance. Ours is a prevalent cohort of DRE and mortality was evaluated after the period of increased mortality risk in an incident cohort with epilepsy (ie, the first 5 years after diagnosis). In our study, we also demonstrate increased mortality. Thus, in this group with a median duration of epilepsy at entry of 25 years, elevated mortality persists. There have been mixed results regarding the SMR in patients with epilepsy of unknown aetiology with some studies demonstrating a significantly increased SMR and others not. We found an elevated SMR within this subgroup that was not statistically significant. We conclude that there is an increased mortality within a prevalence cohort of adult DRE due to those with a known aetiology of epilepsy, even after exclusion of those with a progressive cause. To definitively determine if there is an increased mortality in DRE patients with an unknown aetiology, a larger cohort must be followed.

The closest study for comparison to ours is Mohanraj et al.8 In patients with newly diagnosed epilepsy, case fatality was 8.1% at 6 years, and 7.7% in their cohort with chronic uncontrolled epilepsy. The cumulative probability of death was approximately 8–9% in both groups at 6 years. Our results in a cohort with DRE are similar with a case fatality of 7.6%, and a cumulative probability of death of 8.7% at 6 years despite several differences in our populations. We required patients to have failed at least two AEDs and have a seizure frequency of more than one per month, whereas, Mohanraj et al required patients to have currently uncontrolled seizures of any frequency. They did not require that patients fail a certain number of drugs or have a specific seizure frequency. Therefore, our cohort likely had consistently more severe epilepsy; it was also slightly older. Despite these differences, mortality was comparable, suggesting that the aetiology of epilepsy and continued seizures are the most important determinants of mortality and not the absolute frequency of seizures or the number of AEDs failed.

The causes of death identified in this population were similar to previously described unselected epilepsy cohorts.4 We found that cancer, heart disease, SUDEP, status epilepticus, stroke and neurodegenerative disease were the most common causes of death. In 7 out of the 33 (21%) deaths, the underlying cause of the epilepsy was also the cause of death; for 5 of the 7, this was a brain tumour. Epilepsy-related deaths, including SUDEP, status epilepticus, accidents and suicides, accounted for 30% of the deaths. Five people (1.2%) died of SUDEP in our population, which is probably an underestimate, given that death certificates likely underestimate this cause of death.

Frequently, the treating epileptologist appeared to be unaware of the patient's death. In fact, one-third of the 33 patients who died were believed to be lost to follow-up. Moreover, 12% of patients thought by their doctor to be lost to follow-up were actually deceased. Patients lost to follow-up should, therefore, not be presumed to be alive in studies of DRE. We were able to identify deaths in these patients through the SSDI. Clearly, systematic follow-up is necessary to understand the short-term and long-term consequences of DRE. In this population, loss to follow-up may be more than just a desire for a different opinion.

The two clinical factors associated with an increased risk of death were older age at study entry and ‘symptomatic generalised epilepsy syndrome’ (in pre-2010 terminology). A known aetiology of epilepsy was also associated with a trend toward an increased risk of death. The increased risk of death associated with symptomatic generalised epilepsy is consistent with previous groups showing that children with neurodeficits at birth have a statistically significant SMR.2 ,10 Gender, age of epilepsy onset, and number of AEDs failed were not associated with mortality, providing further evidence that epilepsy syndrome and aetiology are the most important risk factors for death. Even though we did not find an association between number of AEDs failed and overall mortality, previous work has demonstrated that failure of two or more AEDs is associated with death from SUDEP.23 One explanation for these findings is that drug resistance, and not the total number of AEDs failed, is the important determinant of mortality. Alternatively, the clinical factors associated with overall mortality may be different than those for SUDEP.

Our study has some limitations. The lack of a statistically significant SMR for epilepsy of unknown aetiology or for certain predictors of mortality may reflect a relatively small sample size. A further limitation was that both cohorts were followed at tertiary care centres; therefore, the generalisability to other settings is unknown. Those lost to follow-up in our study, 19% of the cohort, may have had a different cumulative probability of death, but we think this unlikely, as the SSDI failed to identify deaths in this group even when checking years after the last reported death. Cause of death was obtained primarily from death certificates, which is not as accurate as autopsy reports.24 However, autopsy is rare in non-forensic deaths in the USA, especially in adult populations.25 ,26 Our study is restricted to the inclusion of a prevalent cohort of DRE rather than an incident DRE cohort. On the other hand, an incident DRE cohort with seizures >1/month would take many years to identify and would require an extremely large sample size. Furthermore, a study of a prevalence cohort provides useful prognostic information for tertiary care epileptologists, whose patients are frequently drug-resistant prior to being seen for the first time. We were also unable to use the ILAE definition of DRE, because this requires detailed information regarding drug failures that was not available in our retrospective medical chart abstraction. Other questions which we could not address could be addressed in an incident DRE cohort, such as the effect of the number of AEDs failed on mortality.

Similar to unselected epilepsy populations, there is increased mortality in DRE, particularly in the subgroup with a known aetiology of epilepsy. Comparing the cumulative incidence and case fatality in our population to those with newly diagnosed or chronic epilepsy, reveals a strikingly similar risk of death in our population, although our population is more chronic. Interestingly, many patients were not known to be deceased by their epileptologist. Future studies of DRE will need to confirm loss to follow-up by first searching death indexes.


View Abstract


  • Contributors BC was involved in the design and analysis of this study and writing of the manuscript. HC was involved in the analysis of the study and revising the manuscript. MS performed statistical analysis of the data. WR was involved in the design of the study. JP helped with the analysis of the study. DCH, WAH and JF were involved in the design and analysis of this study and revising of the manuscript. The statistical analyses were performed by BC, DCH and MS .

  • Competing interests BC reports no disclosures. HC has received research grant funding from Lundbeck and UCB Inc. MS reports no discourses. WR reports no disclosures. JP has received personal compensation from a consulting arrangement with H Lundbeck Inc. He has received grant support from H Lunbeck Inc., Vertex Pharmaceuticals, SK Life Science, Pfizer, Medtronic, Glaxo SmithKline, Eisai, Cyberonics, and Upsher-Smith. DCH serves on the editorial board of Epilepsy and Behavior and Epilepsy Research and as contributing editor to Epilepsy Currents. She consults for the Mount Sinai Medical Center, Injury prevention centre. She received a travel award from GlaxoSmithKline in 2010. In 2012, she participated in advisory boards for UCB, UpsherSmith and Esai. She is funded by grants from CDC, DP002209, PI, 2009–2014; AUCD, RT01, Co-I (PI of Columbia subcontract), 2008–2012; NINDS, NS31146, Co-I (PI of Columbia subcontract), 2007–2014; NINDS, NS043209, Co-I (PI of Columbia subcontract), 2003–2013; CDC, MM1002, Co-I, 2006–2010; NICHD, HD042823, Co-I, 2002–2013; NINDS, 5U01NS04911, Co-I (PI of Columbia subcontract), 2011–2012; NINDS, NS078419, Co-I, 2012–2015; and the Epilepsy Foundation of America 2010–2012. WAH reports that he is a consultant for Neuropace and the University of Kansas. He receives grant support from the CDC and is on the DSMB for Teva. JF has received grant funding from The Milken Foundation, the Epilepsy Therapy Project, and NINDS. She serves as the president of The Epilepsy Study Consortium, a non-profit organisation. NYU receives a fixed amount from the Epilepsy Study Consortium towards JF's salary. The money is for work performed by JF on behalf of The Epilepsy Study Consortium, for consulting and clinical trial related activities. JF receives no personal income for these activities. Within the past year, The Epilepsy Study Consortium received payments from: Aprecia, Avanir, Biotie, Catalyst, Concert, Cyberonics, Eisai Medical Research, Eli Lilly, GlaxoSmithKline, Icagen, Inc. Impax, Johnson & Johnson,LGCH Inc, Mapp Pharmaceuticals, Marinus, Neurelis, Neurotherapeutics, Neuropace, NeuroVista Corporation, Novartis, Ono Pharma USA, Inc., Lundbeck, Pfizer, Sepracor, Sunovion, SK Life Science, Supernus Pharmaceuticals, UCB Inc/Schwarz Pharma, Upsher Smith, Valeant, Vertex, Vivus.

  • Ethics approval IRBs of Columbia and University of Pennsylvania.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.